Building a List of Synonymous Words and Phrases of Japanese Compound Verbs

Building a List of Synonymous Words and Phrases of Japanese Compound Verbs Kyoko Kanzaki and Hitoshi Isahara Toyohashi University of Technology 1-1 Hibarigaoka Tenpaku-cho, Toyohashi, Aichi 441-8580 kanzaki@imc.tut.ac.jp, isahara@tut.jp Abstract We started to construct a database of synonymous expressions of Japanese “Verb + Verb” compounds semi-automatically. Japanese is known to be rich in compound verbs consisting of two verbs joined together. However, we did not have a comprehensive Japanese compound lexicon. Recently a Japanese compound verb lexicon was constructed by the National Institute for Japanese Language and Linguistics(NINJAL)(2013-15). Though it has meanings, example sentences, syntactic patterns and actual sentences from the corpus that they possess, it has no information on relationships with another words, such as synonymous words and phrases. We automatically extracted synonymous expressions of compound verbs from corpus which is “five hundred million Japanese texts gathered from the web” produced by Kawahara et.al. (2006) by using word2vec and cosine similarity and find suitable clusters which correspond to meanings of the compound verbs by using k-means++ and PCA. The automatic extraction from corpus helps humans find not only typical synonyms but also unexpected synonymous words and phrases. Then we manually compile the list of synonymous expressions of Japanese compound verbs by assessing the result and also link it to the “Compound Verb Lexicon” published by NINJAL. Keywords: database, Japanese compound verbs, synonymous words and phrases 1. Introduction In this work, we deal with Japanese verbs and in particular, those of the compound variety. Japanese “Verb+Verb” compounds frequently appear in daily communication which is related to human actions. In terms of morphology, Japanese compounds involving verbs and other predicates are productive and widespread in both syntactic and lexical domains. We have started to construct a database of synonymous expressions of Japanese “Verb + Verb” compounds semi-automatically. Recently the Japanese compound verb lexicon was constructed by the National Institute for Japanese Language and Linguistics (NINJAL) (2013-15). It has meanings, example sentences, syntactic patterns and actual sentences from the corpus that they possess. However, it has no relationship information with another words, such as synonymous words and phrases. We try to detect them automatically as much as possible in order to help humans find not only typical synonyms but also unexpected synonymous words and phrases and manually compile a lexicon of synonymous expressions of Japanese compound verbs. We conducted three actions, 1) extract synonymous expressions (words and phrases) from corpus by using word2vec and cosine similarity measure, 2) classify synonymous expressions into several clusters by using the clustering method k-means++ (Arthur and Vassilvitskii 2009) and 3) find a suitable number of clusters corresponding to the meanings of each compound verb by PCA and compile a list of synonymous expressions and their semantic classes by assessing them manually. 2. Compound verbs that we treat Japanese compound verbs consist of two verbs joined together. The morphological form is a combination of the first verb in an adnominal form and the second verb coming after it, as in hikari (adnominal form)-kagayaku (give.off.light & shine) ‘shine …

Content truncated to 500 words